Logit Averaging

Logit Averaging is a novel yet simple algorithm (LA), helping to learn both local-level logits which come from intra-sessions’ item transitions and global-level logits, which come from gathered logits of related sessions.

Overall Framework of Logit Averaging

Setups

Datasets

The dataset name must be specified in the "--dataset" argument

• Diginetica
• Retailrocket
• Yoochoose 1/64 (using latest 1/64 fraction due to the amount of full dataset)
• Tmall

Train and Test

Run main.py file to train the model. You can configure some training parameters through the command line.

python main.py

Citation

Please cite our paper if you use the code:

@article{yang2022la,
  title={Logit Averaging: Capturing Global Relation for
Session-Based Recommendation},
  author={Heeyoon Yang, Gahyung Kim, Jee-Hyong Lee},
  journal={Applied Science - Special Issue},
  year={2022},
  doi={10.3390/app12094256}
}

Reference

1. NARM: Neural Attentive Session-based Recommendation
2. EOPA of LESSR: Handling Information Loss of Graph Neural Networks for Session-based Recommendation
3. NISER: Normalized Item and Session Representations to Handle Popularity Bias
4. SRGNN: Session-based Recommendation with Graph Neural Network
5. SRSAN: Session-based Recommendation with Self-Attention Networks
6. TAGNN++: Introducing Self-Attention to Target Attentive Graph Neural Networks